An Approach using transformer architecture for emotion recognition through Electrocardiogram Signal (s)

This study proposed an AI-based approach to detect seven emotional states (Happiness, Sadness, Surprise, Anger, Fear, Disgust, Neutral) based on an electrocardiogram (ECG). A well-known three-dimensional model (valence, arousal & dominance), also known as the PAD model, was used to classify the emotional spectrum. We propose a network architecture, Transformer and Temporal Convolution Network, based solely on attention mechanisms, without recurrence and convolution. A comparative analysis between different transfer learning and fine-tuning techniques was then carried out. Three databases were used, starting with the MIT BIH (Massachusetts Institute of Technology and Beth Israel Hospital) database for the characteristics of the recorded signals, and the DREAMER (Dataset for Emotion Analysis using EEG, Physiological and Video Signals) and YAAD (Young Adult Age Dataset) databases for the physiological recordings and subjective ratings of the PAD values. In this paper we address two different problems (heart disease and emotion recognition) using electrocardiogram signals. Evaluation metrics such as Mean Absolute Error and Mean Squared Error were used to assess the performance of the transfer learning models. The overall goal of this study is to analyze and compare the performance of the model and two different problems to understand the emotion in different scenarios. This includes all techniques for automatic evaluation of emotions for applications in marketing, video games, social media, website customization, healthcare, education and other fields.